February 6, 2026
What if nuance is the moat?
By Amias Gerety & Adams Conrad
The whole internet is currently obsessed with OpenClaw, but it’s not an LLM and it doesn’t really have any proprietary AI. Instead, it found a way to make the LLMs feel like they work and, more importantly, feel like they work for you.
The genericization problem
There’s a telling experiment you can try right now. Ask ChatGPT to help write your emails. You’ll notice something odd—a kind of uncanny valley of formality. It’s way more formal than you’d actually be with your team, but somehow less formal than you’d be in certain high-stakes situations. It’s verbose in a way that feels slightly off.
Now extrapolate that to compliance reports. Underwriting memos. Client briefing notes. There’s an old saw that culture is not what we say, but how we do things. Any place there is a “culture,” any place there is a way that a company does things, that is where the nuance can be registered.
And here, the fault of the LLMs becomes clear: LLMs have an incredible capacity for genericization. They smooth out the edges, neutralize the voice, standardize the approach.
The billion-dollar question
The risk of hyperscalers eating everyone’s lunch is very real. Some VCs say “moats don’t matter anymore.” Others claim “capital is the moat—the winner will be determined by depth of balance sheet.”
But what if the moat is something else entirely?
What if the moat is nuance?
Think about employees in your organization who truly understand how things work. The ones who know not just the process, but the unwritten rules. The ones who understand the tone your CEO expects in board materials, the level of detail your compliance team needs, the style of analysis that resonates with your investment committee. These employees are difficult to replace—not because their skills are rare, but because they’ve absorbed the nuance of your organization.
AI will be similar.
QED portfolio company ModelML automates the analytical work that drives much of the day to day for consultants, bankers and investment analysts, combining a set of technically difficult base skills that demand high accuracy (like checking your powerpoint for internal inconsistencies) with a customization interface that allows each client to hold “the way” they prepare decks or client briefs consistent across users.
Two types of steps, one critical interface
Any business process combines deterministic and non-deterministic steps. Deterministic: “What was the funding amount at this company’s last round?” Non-deterministic: “What’s the tone in our meeting notes? Is the sentiment positive or negative?” The real opportunity—and the real moat—lies in managing the interaction between these two types of steps.
For example, Footprint has an AI agent called Percy who has the ability to create dynamic interfaces and also call deterministic steps like, “If you’re uncertain about this, then ask the user for more information or then pull us a score from the credit bureau.”
This is where nuance compounds. The companies that figure out how to tune LLM capabilities to their specific workflows, that build the product engineering to execute consistently, that absorb and operationalize their clients’ nuance—these are the ones building something sticky.
The prompt engineering reality
Despite all the progress in AI, there’s a truth that often gets glossed over in demos:
Give an LLM a generic prompt, even a detailed one that says “do X, then do Y, then do Z,” and it tends to peter out over time. It takes real work to identify when you need to break things down, when a human needs to intervene or how to string together multiple steps. This kind of specific product orientation—the unglamorous work of engineering reliable workflows—is where niche players can win against the hyperscalers. It’s not sexy, but it’s defensible.
The system of record imperative
This puts enormous pressure on where AI sits in the organization. If nuance is the moat, then building systems of record and daily-use workstations becomes critical. You need to be in the flow of work, capturing context as it happens, learning the patterns of how decisions actually get made. Because in the end, an employee who deeply understands your business is hard to replace. AI that deeply understands your business will be, too.
The question for builders and investors isn’t whether AI will transform your industry—it will. The question is: who will capture and defend the nuance that makes your organization unique?
By Amias Gerety & Adams Conrad
The whole internet is currently obsessed with OpenClaw, but it’s not an LLM and it doesn’t really have any proprietary AI. Instead, it found a way to make the LLMs feel like they work and, more importantly, feel like they work for you.
The genericization problem
There’s a telling experiment you can try right now. Ask ChatGPT to help write your emails. You’ll notice something odd—a kind of uncanny valley of formality. It’s way more formal than you’d actually be with your team, but somehow less formal than you’d be in certain high-stakes situations. It’s verbose in a way that feels slightly off.
Now extrapolate that to compliance reports. Underwriting memos. Client briefing notes. There’s an old saw that culture is not what we say, but how we do things. Any place there is a “culture,” any place there is a way that a company does things, that is where the nuance can be registered.
And here, the fault of the LLMs becomes clear: LLMs have an incredible capacity for genericization. They smooth out the edges, neutralize the voice, standardize the approach.
The billion-dollar question
The risk of hyperscalers eating everyone’s lunch is very real. Some VCs say “moats don’t matter anymore.” Others claim “capital is the moat—the winner will be determined by depth of balance sheet.”
But what if the moat is something else entirely?
What if the moat is nuance?
Think about employees in your organization who truly understand how things work. The ones who know not just the process, but the unwritten rules. The ones who understand the tone your CEO expects in board materials, the level of detail your compliance team needs, the style of analysis that resonates with your investment committee. These employees are difficult to replace—not because their skills are rare, but because they’ve absorbed the nuance of your organization.
AI will be similar.
QED portfolio company ModelML automates the analytical work that drives much of the day to day for consultants, bankers and investment analysts, combining a set of technically difficult base skills that demand high accuracy (like checking your powerpoint for internal inconsistencies) with a customization interface that allows each client to hold “the way” they prepare decks or client briefs consistent across users.
Two types of steps, one critical interface
Any business process combines deterministic and non-deterministic steps. Deterministic: “What was the funding amount at this company’s last round?” Non-deterministic: “What’s the tone in our meeting notes? Is the sentiment positive or negative?” The real opportunity—and the real moat—lies in managing the interaction between these two types of steps.
For example, Footprint has an AI agent called Percy who has the ability to create dynamic interfaces and also call deterministic steps like, “If you’re uncertain about this, then ask the user for more information or then pull us a score from the credit bureau.”
This is where nuance compounds. The companies that figure out how to tune LLM capabilities to their specific workflows, that build the product engineering to execute consistently, that absorb and operationalize their clients’ nuance—these are the ones building something sticky.
The prompt engineering reality
Despite all the progress in AI, there’s a truth that often gets glossed over in demos:
Give an LLM a generic prompt, even a detailed one that says “do X, then do Y, then do Z,” and it tends to peter out over time. It takes real work to identify when you need to break things down, when a human needs to intervene or how to string together multiple steps. This kind of specific product orientation—the unglamorous work of engineering reliable workflows—is where niche players can win against the hyperscalers. It’s not sexy, but it’s defensible.
The system of record imperative
This puts enormous pressure on where AI sits in the organization. If nuance is the moat, then building systems of record and daily-use workstations becomes critical. You need to be in the flow of work, capturing context as it happens, learning the patterns of how decisions actually get made. Because in the end, an employee who deeply understands your business is hard to replace. AI that deeply understands your business will be, too.
The question for builders and investors isn’t whether AI will transform your industry—it will. The question is: who will capture and defend the nuance that makes your organization unique?